Nanbeige4.3-33B vs Qwen3.5-397B-A17B

Nanbeige4.3-33B

A dense 33B Llama-architecture cousin for long agentic coding and office tasks, with per-request thinking, XML tool calls that stock vLLM and SGLang parse, and 512K of context. It needs about 96 GB at BF16, as Nanbeige ships no quantised build. Its agentic results so far are Nanbeige's own, with no safety results, training data notes or licence file.

Qwen3.5-397B-A17B

Alibaba's biggest open Qwen3.5, a cousin for hard coding, agent and vision work. Even at 4 bits it needs about 270 GB, so most agents will call it through Alibaba's Model Studio or one of the many providers that host it. The main risk is silence, with no safety results, no agent guidance and template reports without a reply.

Nanbeige4.3-33B

1. Serve with vllm serve Nanbeige/Nanbeige4.3-33B --enable-auto-tool-choice --reasoning-parser qwen3 --tool-call-parser qwen3_coder, as the card says, or SGLang with the same two parsers
2. Keep tool_call_format at its default xml, since qwen3_coder parses that format; the json option writes Hermes-style JSON calls, which need a different parser such as vLLM's hermes
3. Pass chat_template_kwargs {"preserve_thinking": true} for multi-turn tool use and coding, and {"enable_thinking": false} to skip reasoning, as the card recommends
4. Use temperature 1.0, top_p 0.95 and up to 65,536 new tokens for agent work, and temperature 0.6 for reasoning and chat, per the card
5. Send your own system prompt, since without one the template injects a Chinese identity line, or a Chinese tool-expert prompt when tools are passed
6. Send one system message at the start only, as the template raises an error for a later one
7. In plain Transformers pass use_cache=True to generate, since config.json and generation_config.json set it false; trust_remote_code isn't needed

Qwen3.5-397B-A17B

1. The card's server command is vllm serve Qwen/Qwen3.5-397B-A17B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder, and the FP8 build halves the memory
2. It doesn't fit our 128 GB reference machine at any precision, so call a hosted endpoint instead (Model Studio's Qwen3.5-Plus, or a provider serving Qwen/Qwen3.5-397B-A17B)
3. On Model Studio pass "enable_thinking": false directly, while vLLM and SGLang take "chat_template_kwargs": {"enable_thinking": false}
4. Sampling per the card is temperature 0.6, top_p 0.95 and top_k 20 when thinking, and temperature 0.7, top_p 0.8, top_k 20 and presence_penalty 1.5 without
5. For multi-token prediction in vLLM add --speculative-config '{"method":"qwen3_next_mtp","num_speculative_tokens":2}'
6. Keep the context at 128K or more if you rely on thinking, and turn on YaRN only past 262,144 tokens
7. Validate any JSON it returns, since structured output isn't documented